Action valuation of on- and off-ball soccer players based on multi-agent deep reinforcement learning.
Hiroshi NakaharaKazushi TsutsuiKazuya TakedaKeisuke FujiiPublished in: CoRR (2023)
Keyphrases
- reinforcement learning
- multi agent
- action selection
- soccer games
- soccer game
- robot soccer
- robocup soccer
- function approximation
- action space
- state action
- learning agents
- event detection
- multi agent reinforcement learning
- cooperative
- game theory
- reward shaping
- markov decision processes
- multi agent systems
- extensive form games
- optimal policy
- reinforcement learning algorithms
- multi agent environments
- model free
- single agent
- reinforcement learning agents
- transition model
- partially observable domains
- continuous state
- agent learns
- fitted q iteration
- reinforcement learning methods
- soccer video
- temporal difference
- multiagent systems
- vision system
- state space
- decision making
- sensory inputs
- optimal control
- transfer learning
- intelligent agents
- machine learning
- partial observability
- function approximators
- multiple agents
- learning process